1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
11 #include <rte_ethdev_driver.h>
12 #include <rte_ethdev_pci.h>
13 #include <rte_memcpy.h>
14 #include <rte_string_fns.h>
15 #include <rte_memzone.h>
16 #include <rte_malloc.h>
17 #include <rte_branch_prediction.h>
19 #include <rte_bus_pci.h>
20 #include <rte_ether.h>
23 #include <rte_common.h>
24 #include <rte_errno.h>
25 #include <rte_cpuflags.h>
27 #include <rte_memory.h>
30 #include <rte_cycles.h>
32 #include "virtio_ethdev.h"
33 #include "virtio_pci.h"
34 #include "virtio_logs.h"
35 #include "virtqueue.h"
36 #include "virtio_rxtx.h"
38 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
39 static int virtio_dev_configure(struct rte_eth_dev *dev);
40 static int virtio_dev_start(struct rte_eth_dev *dev);
41 static void virtio_dev_stop(struct rte_eth_dev *dev);
42 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
43 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
44 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
45 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
46 static void virtio_dev_info_get(struct rte_eth_dev *dev,
47 struct rte_eth_dev_info *dev_info);
48 static int virtio_dev_link_update(struct rte_eth_dev *dev,
49 int wait_to_complete);
50 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
52 static void virtio_set_hwaddr(struct virtio_hw *hw);
53 static void virtio_get_hwaddr(struct virtio_hw *hw);
55 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
56 struct rte_eth_stats *stats);
57 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
58 struct rte_eth_xstat *xstats, unsigned n);
59 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
60 struct rte_eth_xstat_name *xstats_names,
62 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
63 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
64 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
65 uint16_t vlan_id, int on);
66 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
67 struct ether_addr *mac_addr,
68 uint32_t index, uint32_t vmdq);
69 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
70 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
71 struct ether_addr *mac_addr);
73 static int virtio_intr_enable(struct rte_eth_dev *dev);
74 static int virtio_intr_disable(struct rte_eth_dev *dev);
76 static int virtio_dev_queue_stats_mapping_set(
77 struct rte_eth_dev *eth_dev,
82 int virtio_logtype_init;
83 int virtio_logtype_driver;
85 static void virtio_notify_peers(struct rte_eth_dev *dev);
86 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
89 * The set of PCI devices this driver supports
91 static const struct rte_pci_id pci_id_virtio_map[] = {
92 { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
93 { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
94 { .vendor_id = 0, /* sentinel */ },
97 struct rte_virtio_xstats_name_off {
98 char name[RTE_ETH_XSTATS_NAME_SIZE];
102 /* [rt]x_qX_ is prepended to the name string here */
103 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
104 {"good_packets", offsetof(struct virtnet_rx, stats.packets)},
105 {"good_bytes", offsetof(struct virtnet_rx, stats.bytes)},
106 {"errors", offsetof(struct virtnet_rx, stats.errors)},
107 {"multicast_packets", offsetof(struct virtnet_rx, stats.multicast)},
108 {"broadcast_packets", offsetof(struct virtnet_rx, stats.broadcast)},
109 {"undersize_packets", offsetof(struct virtnet_rx, stats.size_bins[0])},
110 {"size_64_packets", offsetof(struct virtnet_rx, stats.size_bins[1])},
111 {"size_65_127_packets", offsetof(struct virtnet_rx, stats.size_bins[2])},
112 {"size_128_255_packets", offsetof(struct virtnet_rx, stats.size_bins[3])},
113 {"size_256_511_packets", offsetof(struct virtnet_rx, stats.size_bins[4])},
114 {"size_512_1023_packets", offsetof(struct virtnet_rx, stats.size_bins[5])},
115 {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
116 {"size_1519_max_packets", offsetof(struct virtnet_rx, stats.size_bins[7])},
119 /* [rt]x_qX_ is prepended to the name string here */
120 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
121 {"good_packets", offsetof(struct virtnet_tx, stats.packets)},
122 {"good_bytes", offsetof(struct virtnet_tx, stats.bytes)},
123 {"errors", offsetof(struct virtnet_tx, stats.errors)},
124 {"multicast_packets", offsetof(struct virtnet_tx, stats.multicast)},
125 {"broadcast_packets", offsetof(struct virtnet_tx, stats.broadcast)},
126 {"undersize_packets", offsetof(struct virtnet_tx, stats.size_bins[0])},
127 {"size_64_packets", offsetof(struct virtnet_tx, stats.size_bins[1])},
128 {"size_65_127_packets", offsetof(struct virtnet_tx, stats.size_bins[2])},
129 {"size_128_255_packets", offsetof(struct virtnet_tx, stats.size_bins[3])},
130 {"size_256_511_packets", offsetof(struct virtnet_tx, stats.size_bins[4])},
131 {"size_512_1023_packets", offsetof(struct virtnet_tx, stats.size_bins[5])},
132 {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
133 {"size_1519_max_packets", offsetof(struct virtnet_tx, stats.size_bins[7])},
136 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
137 sizeof(rte_virtio_rxq_stat_strings[0]))
138 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
139 sizeof(rte_virtio_txq_stat_strings[0]))
141 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
144 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
145 int *dlen, int pkt_num)
149 virtio_net_ctrl_ack status = ~0;
150 struct virtio_pmd_ctrl *result;
151 struct virtqueue *vq;
153 ctrl->status = status;
155 if (!cvq || !cvq->vq) {
156 PMD_INIT_LOG(ERR, "Control queue is not supported.");
160 rte_spinlock_lock(&cvq->lock);
162 head = vq->vq_desc_head_idx;
164 PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
165 "vq->hw->cvq = %p vq = %p",
166 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
168 if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
169 rte_spinlock_unlock(&cvq->lock);
173 memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
174 sizeof(struct virtio_pmd_ctrl));
177 * Format is enforced in qemu code:
178 * One TX packet for header;
179 * At least one TX packet per argument;
180 * One RX packet for ACK.
182 vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
183 vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
184 vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
186 i = vq->vq_ring.desc[head].next;
188 for (k = 0; k < pkt_num; k++) {
189 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
190 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
191 + sizeof(struct virtio_net_ctrl_hdr)
192 + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
193 vq->vq_ring.desc[i].len = dlen[k];
196 i = vq->vq_ring.desc[i].next;
199 vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
200 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
201 + sizeof(struct virtio_net_ctrl_hdr);
202 vq->vq_ring.desc[i].len = sizeof(ctrl->status);
205 vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
207 vq_update_avail_ring(vq, head);
208 vq_update_avail_idx(vq);
210 PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
212 virtqueue_notify(vq);
215 while (VIRTQUEUE_NUSED(vq) == 0) {
220 while (VIRTQUEUE_NUSED(vq)) {
221 uint32_t idx, desc_idx, used_idx;
222 struct vring_used_elem *uep;
224 used_idx = (uint32_t)(vq->vq_used_cons_idx
225 & (vq->vq_nentries - 1));
226 uep = &vq->vq_ring.used->ring[used_idx];
227 idx = (uint32_t) uep->id;
230 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
231 desc_idx = vq->vq_ring.desc[desc_idx].next;
235 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
236 vq->vq_desc_head_idx = idx;
238 vq->vq_used_cons_idx++;
242 PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
243 vq->vq_free_cnt, vq->vq_desc_head_idx);
245 result = cvq->virtio_net_hdr_mz->addr;
247 rte_spinlock_unlock(&cvq->lock);
248 return result->status;
252 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
254 struct virtio_hw *hw = dev->data->dev_private;
255 struct virtio_pmd_ctrl ctrl;
259 ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
260 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
261 memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
263 dlen[0] = sizeof(uint16_t);
265 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
267 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
268 "failed, this is too late now...");
276 virtio_dev_queue_release(void *queue __rte_unused)
282 virtio_get_nr_vq(struct virtio_hw *hw)
284 uint16_t nr_vq = hw->max_queue_pairs * 2;
286 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
293 virtio_init_vring(struct virtqueue *vq)
295 int size = vq->vq_nentries;
296 struct vring *vr = &vq->vq_ring;
297 uint8_t *ring_mem = vq->vq_ring_virt_mem;
299 PMD_INIT_FUNC_TRACE();
302 * Reinitialise since virtio port might have been stopped and restarted
304 memset(ring_mem, 0, vq->vq_ring_size);
305 vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
306 vq->vq_used_cons_idx = 0;
307 vq->vq_desc_head_idx = 0;
308 vq->vq_avail_idx = 0;
309 vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
310 vq->vq_free_cnt = vq->vq_nentries;
311 memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
313 vring_desc_init(vr->desc, size);
316 * Disable device(host) interrupting guest
318 virtqueue_disable_intr(vq);
322 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
324 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
325 char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
326 const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
327 unsigned int vq_size, size;
328 struct virtio_hw *hw = dev->data->dev_private;
329 struct virtnet_rx *rxvq = NULL;
330 struct virtnet_tx *txvq = NULL;
331 struct virtnet_ctl *cvq = NULL;
332 struct virtqueue *vq;
333 size_t sz_hdr_mz = 0;
334 void *sw_ring = NULL;
335 int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
338 PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
341 * Read the virtqueue size from the Queue Size field
342 * Always power of 2 and if 0 virtqueue does not exist
344 vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
345 PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
347 PMD_INIT_LOG(ERR, "virtqueue does not exist");
351 if (!rte_is_power_of_2(vq_size)) {
352 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
356 snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
357 dev->data->port_id, vtpci_queue_idx);
359 size = RTE_ALIGN_CEIL(sizeof(*vq) +
360 vq_size * sizeof(struct vq_desc_extra),
361 RTE_CACHE_LINE_SIZE);
362 if (queue_type == VTNET_TQ) {
364 * For each xmit packet, allocate a virtio_net_hdr
365 * and indirect ring elements
367 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
368 } else if (queue_type == VTNET_CQ) {
369 /* Allocate a page for control vq command, data and status */
370 sz_hdr_mz = PAGE_SIZE;
373 vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
376 PMD_INIT_LOG(ERR, "can not allocate vq");
379 hw->vqs[vtpci_queue_idx] = vq;
382 vq->vq_queue_index = vtpci_queue_idx;
383 vq->vq_nentries = vq_size;
386 * Reserve a memzone for vring elements
388 size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
389 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
390 PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
391 size, vq->vq_ring_size);
393 mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
394 SOCKET_ID_ANY, RTE_MEMZONE_IOVA_CONTIG,
395 VIRTIO_PCI_VRING_ALIGN);
397 if (rte_errno == EEXIST)
398 mz = rte_memzone_lookup(vq_name);
405 memset(mz->addr, 0, mz->len);
407 vq->vq_ring_mem = mz->iova;
408 vq->vq_ring_virt_mem = mz->addr;
409 PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%" PRIx64,
411 PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
412 (uint64_t)(uintptr_t)mz->addr);
414 virtio_init_vring(vq);
417 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
418 dev->data->port_id, vtpci_queue_idx);
419 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
420 SOCKET_ID_ANY, RTE_MEMZONE_IOVA_CONTIG,
421 RTE_CACHE_LINE_SIZE);
422 if (hdr_mz == NULL) {
423 if (rte_errno == EEXIST)
424 hdr_mz = rte_memzone_lookup(vq_hdr_name);
425 if (hdr_mz == NULL) {
432 if (queue_type == VTNET_RQ) {
433 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
434 sizeof(vq->sw_ring[0]);
436 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
437 RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
439 PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
444 vq->sw_ring = sw_ring;
447 rxvq->port_id = dev->data->port_id;
449 } else if (queue_type == VTNET_TQ) {
452 txvq->port_id = dev->data->port_id;
454 txvq->virtio_net_hdr_mz = hdr_mz;
455 txvq->virtio_net_hdr_mem = hdr_mz->iova;
456 } else if (queue_type == VTNET_CQ) {
460 cvq->virtio_net_hdr_mz = hdr_mz;
461 cvq->virtio_net_hdr_mem = hdr_mz->iova;
462 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
467 /* For virtio_user case (that is when hw->dev is NULL), we use
468 * virtual address. And we need properly set _offset_, please see
469 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
471 if (!hw->virtio_user_dev)
472 vq->offset = offsetof(struct rte_mbuf, buf_iova);
474 vq->vq_ring_mem = (uintptr_t)mz->addr;
475 vq->offset = offsetof(struct rte_mbuf, buf_addr);
476 if (queue_type == VTNET_TQ)
477 txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
478 else if (queue_type == VTNET_CQ)
479 cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
482 if (queue_type == VTNET_TQ) {
483 struct virtio_tx_region *txr;
487 memset(txr, 0, vq_size * sizeof(*txr));
488 for (i = 0; i < vq_size; i++) {
489 struct vring_desc *start_dp = txr[i].tx_indir;
491 vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
493 /* first indirect descriptor is always the tx header */
494 start_dp->addr = txvq->virtio_net_hdr_mem
496 + offsetof(struct virtio_tx_region, tx_hdr);
498 start_dp->len = hw->vtnet_hdr_size;
499 start_dp->flags = VRING_DESC_F_NEXT;
503 if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
504 PMD_INIT_LOG(ERR, "setup_queue failed");
512 rte_memzone_free(hdr_mz);
513 rte_memzone_free(mz);
520 virtio_free_queues(struct virtio_hw *hw)
522 uint16_t nr_vq = virtio_get_nr_vq(hw);
523 struct virtqueue *vq;
530 for (i = 0; i < nr_vq; i++) {
535 queue_type = virtio_get_queue_type(hw, i);
536 if (queue_type == VTNET_RQ) {
537 rte_free(vq->sw_ring);
538 rte_memzone_free(vq->rxq.mz);
539 } else if (queue_type == VTNET_TQ) {
540 rte_memzone_free(vq->txq.mz);
541 rte_memzone_free(vq->txq.virtio_net_hdr_mz);
543 rte_memzone_free(vq->cq.mz);
544 rte_memzone_free(vq->cq.virtio_net_hdr_mz);
556 virtio_alloc_queues(struct rte_eth_dev *dev)
558 struct virtio_hw *hw = dev->data->dev_private;
559 uint16_t nr_vq = virtio_get_nr_vq(hw);
563 hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
565 PMD_INIT_LOG(ERR, "failed to allocate vqs");
569 for (i = 0; i < nr_vq; i++) {
570 ret = virtio_init_queue(dev, i);
572 virtio_free_queues(hw);
580 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
583 virtio_dev_close(struct rte_eth_dev *dev)
585 struct virtio_hw *hw = dev->data->dev_private;
586 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
588 PMD_INIT_LOG(DEBUG, "virtio_dev_close");
591 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
592 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
594 virtio_queues_unbind_intr(dev);
596 if (intr_conf->lsc || intr_conf->rxq) {
597 virtio_intr_disable(dev);
598 rte_intr_efd_disable(dev->intr_handle);
599 rte_free(dev->intr_handle->intr_vec);
600 dev->intr_handle->intr_vec = NULL;
604 virtio_dev_free_mbufs(dev);
605 virtio_free_queues(hw);
609 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
611 struct virtio_hw *hw = dev->data->dev_private;
612 struct virtio_pmd_ctrl ctrl;
616 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
617 PMD_INIT_LOG(INFO, "host does not support rx control");
621 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
622 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
626 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
628 PMD_INIT_LOG(ERR, "Failed to enable promisc");
632 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
634 struct virtio_hw *hw = dev->data->dev_private;
635 struct virtio_pmd_ctrl ctrl;
639 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
640 PMD_INIT_LOG(INFO, "host does not support rx control");
644 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
645 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
649 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
651 PMD_INIT_LOG(ERR, "Failed to disable promisc");
655 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
657 struct virtio_hw *hw = dev->data->dev_private;
658 struct virtio_pmd_ctrl ctrl;
662 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
663 PMD_INIT_LOG(INFO, "host does not support rx control");
667 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
668 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
672 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
674 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
678 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
680 struct virtio_hw *hw = dev->data->dev_private;
681 struct virtio_pmd_ctrl ctrl;
685 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
686 PMD_INIT_LOG(INFO, "host does not support rx control");
690 ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
691 ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
695 ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
697 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
700 #define VLAN_TAG_LEN 4 /* 802.3ac tag (not DMA'd) */
702 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
704 struct virtio_hw *hw = dev->data->dev_private;
705 uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
707 uint32_t frame_size = mtu + ether_hdr_len;
708 uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
710 max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
712 if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
713 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
714 ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
721 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
723 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
724 struct virtqueue *vq = rxvq->vq;
726 virtqueue_enable_intr(vq);
731 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
733 struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
734 struct virtqueue *vq = rxvq->vq;
736 virtqueue_disable_intr(vq);
741 * dev_ops for virtio, bare necessities for basic operation
743 static const struct eth_dev_ops virtio_eth_dev_ops = {
744 .dev_configure = virtio_dev_configure,
745 .dev_start = virtio_dev_start,
746 .dev_stop = virtio_dev_stop,
747 .dev_close = virtio_dev_close,
748 .promiscuous_enable = virtio_dev_promiscuous_enable,
749 .promiscuous_disable = virtio_dev_promiscuous_disable,
750 .allmulticast_enable = virtio_dev_allmulticast_enable,
751 .allmulticast_disable = virtio_dev_allmulticast_disable,
752 .mtu_set = virtio_mtu_set,
753 .dev_infos_get = virtio_dev_info_get,
754 .stats_get = virtio_dev_stats_get,
755 .xstats_get = virtio_dev_xstats_get,
756 .xstats_get_names = virtio_dev_xstats_get_names,
757 .stats_reset = virtio_dev_stats_reset,
758 .xstats_reset = virtio_dev_stats_reset,
759 .link_update = virtio_dev_link_update,
760 .vlan_offload_set = virtio_dev_vlan_offload_set,
761 .rx_queue_setup = virtio_dev_rx_queue_setup,
762 .rx_queue_intr_enable = virtio_dev_rx_queue_intr_enable,
763 .rx_queue_intr_disable = virtio_dev_rx_queue_intr_disable,
764 .rx_queue_release = virtio_dev_queue_release,
765 .rx_descriptor_done = virtio_dev_rx_queue_done,
766 .tx_queue_setup = virtio_dev_tx_queue_setup,
767 .tx_queue_release = virtio_dev_queue_release,
768 /* collect stats per queue */
769 .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
770 .vlan_filter_set = virtio_vlan_filter_set,
771 .mac_addr_add = virtio_mac_addr_add,
772 .mac_addr_remove = virtio_mac_addr_remove,
773 .mac_addr_set = virtio_mac_addr_set,
777 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
781 for (i = 0; i < dev->data->nb_tx_queues; i++) {
782 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
786 stats->opackets += txvq->stats.packets;
787 stats->obytes += txvq->stats.bytes;
788 stats->oerrors += txvq->stats.errors;
790 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
791 stats->q_opackets[i] = txvq->stats.packets;
792 stats->q_obytes[i] = txvq->stats.bytes;
796 for (i = 0; i < dev->data->nb_rx_queues; i++) {
797 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
801 stats->ipackets += rxvq->stats.packets;
802 stats->ibytes += rxvq->stats.bytes;
803 stats->ierrors += rxvq->stats.errors;
805 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
806 stats->q_ipackets[i] = rxvq->stats.packets;
807 stats->q_ibytes[i] = rxvq->stats.bytes;
811 stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
814 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
815 struct rte_eth_xstat_name *xstats_names,
816 __rte_unused unsigned limit)
822 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
823 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
825 if (xstats_names != NULL) {
826 /* Note: limit checked in rte_eth_xstats_names() */
828 for (i = 0; i < dev->data->nb_rx_queues; i++) {
829 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
832 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
833 snprintf(xstats_names[count].name,
834 sizeof(xstats_names[count].name),
836 rte_virtio_rxq_stat_strings[t].name);
841 for (i = 0; i < dev->data->nb_tx_queues; i++) {
842 struct virtnet_tx *txvq = dev->data->tx_queues[i];
845 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
846 snprintf(xstats_names[count].name,
847 sizeof(xstats_names[count].name),
849 rte_virtio_txq_stat_strings[t].name);
859 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
865 unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
866 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
871 for (i = 0; i < dev->data->nb_rx_queues; i++) {
872 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
879 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
880 xstats[count].value = *(uint64_t *)(((char *)rxvq) +
881 rte_virtio_rxq_stat_strings[t].offset);
882 xstats[count].id = count;
887 for (i = 0; i < dev->data->nb_tx_queues; i++) {
888 struct virtnet_tx *txvq = dev->data->tx_queues[i];
895 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
896 xstats[count].value = *(uint64_t *)(((char *)txvq) +
897 rte_virtio_txq_stat_strings[t].offset);
898 xstats[count].id = count;
907 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
909 virtio_update_stats(dev, stats);
915 virtio_dev_stats_reset(struct rte_eth_dev *dev)
919 for (i = 0; i < dev->data->nb_tx_queues; i++) {
920 struct virtnet_tx *txvq = dev->data->tx_queues[i];
924 txvq->stats.packets = 0;
925 txvq->stats.bytes = 0;
926 txvq->stats.errors = 0;
927 txvq->stats.multicast = 0;
928 txvq->stats.broadcast = 0;
929 memset(txvq->stats.size_bins, 0,
930 sizeof(txvq->stats.size_bins[0]) * 8);
933 for (i = 0; i < dev->data->nb_rx_queues; i++) {
934 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
938 rxvq->stats.packets = 0;
939 rxvq->stats.bytes = 0;
940 rxvq->stats.errors = 0;
941 rxvq->stats.multicast = 0;
942 rxvq->stats.broadcast = 0;
943 memset(rxvq->stats.size_bins, 0,
944 sizeof(rxvq->stats.size_bins[0]) * 8);
949 virtio_set_hwaddr(struct virtio_hw *hw)
951 vtpci_write_dev_config(hw,
952 offsetof(struct virtio_net_config, mac),
953 &hw->mac_addr, ETHER_ADDR_LEN);
957 virtio_get_hwaddr(struct virtio_hw *hw)
959 if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
960 vtpci_read_dev_config(hw,
961 offsetof(struct virtio_net_config, mac),
962 &hw->mac_addr, ETHER_ADDR_LEN);
964 eth_random_addr(&hw->mac_addr[0]);
965 virtio_set_hwaddr(hw);
970 virtio_mac_table_set(struct virtio_hw *hw,
971 const struct virtio_net_ctrl_mac *uc,
972 const struct virtio_net_ctrl_mac *mc)
974 struct virtio_pmd_ctrl ctrl;
977 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
978 PMD_DRV_LOG(INFO, "host does not support mac table");
982 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
983 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
985 len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
986 memcpy(ctrl.data, uc, len[0]);
988 len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
989 memcpy(ctrl.data + len[0], mc, len[1]);
991 err = virtio_send_command(hw->cvq, &ctrl, len, 2);
993 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
998 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
999 uint32_t index, uint32_t vmdq __rte_unused)
1001 struct virtio_hw *hw = dev->data->dev_private;
1002 const struct ether_addr *addrs = dev->data->mac_addrs;
1004 struct virtio_net_ctrl_mac *uc, *mc;
1006 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1007 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1011 uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1013 mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1016 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1017 const struct ether_addr *addr
1018 = (i == index) ? mac_addr : addrs + i;
1019 struct virtio_net_ctrl_mac *tbl
1020 = is_multicast_ether_addr(addr) ? mc : uc;
1022 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1025 return virtio_mac_table_set(hw, uc, mc);
1029 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1031 struct virtio_hw *hw = dev->data->dev_private;
1032 struct ether_addr *addrs = dev->data->mac_addrs;
1033 struct virtio_net_ctrl_mac *uc, *mc;
1036 if (index >= VIRTIO_MAX_MAC_ADDRS) {
1037 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1041 uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1043 mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1046 for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1047 struct virtio_net_ctrl_mac *tbl;
1049 if (i == index || is_zero_ether_addr(addrs + i))
1052 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1053 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1056 virtio_mac_table_set(hw, uc, mc);
1060 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1062 struct virtio_hw *hw = dev->data->dev_private;
1064 memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1066 /* Use atomic update if available */
1067 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1068 struct virtio_pmd_ctrl ctrl;
1069 int len = ETHER_ADDR_LEN;
1071 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1072 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1074 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1075 virtio_send_command(hw->cvq, &ctrl, &len, 1);
1076 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1077 virtio_set_hwaddr(hw);
1081 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1083 struct virtio_hw *hw = dev->data->dev_private;
1084 struct virtio_pmd_ctrl ctrl;
1087 if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1090 ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1091 ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1092 memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1093 len = sizeof(vlan_id);
1095 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1099 virtio_intr_enable(struct rte_eth_dev *dev)
1101 struct virtio_hw *hw = dev->data->dev_private;
1103 if (rte_intr_enable(dev->intr_handle) < 0)
1106 if (!hw->virtio_user_dev)
1107 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1113 virtio_intr_disable(struct rte_eth_dev *dev)
1115 struct virtio_hw *hw = dev->data->dev_private;
1117 if (rte_intr_disable(dev->intr_handle) < 0)
1120 if (!hw->virtio_user_dev)
1121 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1127 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1129 uint64_t host_features;
1131 /* Prepare guest_features: feature that driver wants to support */
1132 PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1135 /* Read device(host) feature bits */
1136 host_features = VTPCI_OPS(hw)->get_features(hw);
1137 PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1140 /* If supported, ensure MTU value is valid before acknowledging it. */
1141 if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1142 struct virtio_net_config config;
1144 vtpci_read_dev_config(hw,
1145 offsetof(struct virtio_net_config, mtu),
1146 &config.mtu, sizeof(config.mtu));
1148 if (config.mtu < ETHER_MIN_MTU)
1149 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1153 * Negotiate features: Subset of device feature bits are written back
1154 * guest feature bits.
1156 hw->guest_features = req_features;
1157 hw->guest_features = vtpci_negotiate_features(hw, host_features);
1158 PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1159 hw->guest_features);
1162 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1164 "VIRTIO_F_VERSION_1 features is not enabled.");
1167 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1168 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1170 "failed to set FEATURES_OK status!");
1175 hw->req_guest_features = req_features;
1181 virtio_dev_pause(struct rte_eth_dev *dev)
1183 struct virtio_hw *hw = dev->data->dev_private;
1185 rte_spinlock_lock(&hw->state_lock);
1187 if (hw->started == 0) {
1188 /* Device is just stopped. */
1189 rte_spinlock_unlock(&hw->state_lock);
1194 * Prevent the worker threads from touching queues to avoid contention,
1195 * 1 ms should be enough for the ongoing Tx function to finish.
1202 * Recover hw state to let the worker threads continue.
1205 virtio_dev_resume(struct rte_eth_dev *dev)
1207 struct virtio_hw *hw = dev->data->dev_private;
1210 rte_spinlock_unlock(&hw->state_lock);
1214 * Should be called only after device is paused.
1217 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1220 struct virtio_hw *hw = dev->data->dev_private;
1221 struct virtnet_tx *txvq = dev->data->tx_queues[0];
1224 hw->inject_pkts = tx_pkts;
1225 ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1226 hw->inject_pkts = NULL;
1232 virtio_notify_peers(struct rte_eth_dev *dev)
1234 struct virtio_hw *hw = dev->data->dev_private;
1235 struct virtnet_rx *rxvq;
1236 struct rte_mbuf *rarp_mbuf;
1238 if (!dev->data->rx_queues)
1241 rxvq = dev->data->rx_queues[0];
1242 rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1243 (struct ether_addr *)hw->mac_addr);
1244 if (rarp_mbuf == NULL) {
1245 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1249 /* If virtio port just stopped, no need to send RARP */
1250 if (virtio_dev_pause(dev) < 0) {
1251 rte_pktmbuf_free(rarp_mbuf);
1255 virtio_inject_pkts(dev, &rarp_mbuf, 1);
1256 virtio_dev_resume(dev);
1260 virtio_ack_link_announce(struct rte_eth_dev *dev)
1262 struct virtio_hw *hw = dev->data->dev_private;
1263 struct virtio_pmd_ctrl ctrl;
1265 ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1266 ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1268 virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1272 * Process virtio config changed interrupt. Call the callback
1273 * if link state changed, generate gratuitous RARP packet if
1274 * the status indicates an ANNOUNCE.
1277 virtio_interrupt_handler(void *param)
1279 struct rte_eth_dev *dev = param;
1280 struct virtio_hw *hw = dev->data->dev_private;
1283 /* Read interrupt status which clears interrupt */
1284 isr = vtpci_isr(hw);
1285 PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1287 if (virtio_intr_enable(dev) < 0)
1288 PMD_DRV_LOG(ERR, "interrupt enable failed");
1290 if (isr & VIRTIO_PCI_ISR_CONFIG) {
1291 if (virtio_dev_link_update(dev, 0) == 0)
1292 _rte_eth_dev_callback_process(dev,
1293 RTE_ETH_EVENT_INTR_LSC,
1297 if (isr & VIRTIO_NET_S_ANNOUNCE) {
1298 virtio_notify_peers(dev);
1300 virtio_ack_link_announce(dev);
1304 /* set rx and tx handlers according to what is supported */
1306 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1308 struct virtio_hw *hw = eth_dev->data->dev_private;
1310 if (hw->use_simple_rx) {
1311 PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1312 eth_dev->data->port_id);
1313 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1314 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1316 "virtio: using mergeable buffer Rx path on port %u",
1317 eth_dev->data->port_id);
1318 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1320 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1321 eth_dev->data->port_id);
1322 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1325 if (hw->use_simple_tx) {
1326 PMD_INIT_LOG(INFO, "virtio: using simple Tx path on port %u",
1327 eth_dev->data->port_id);
1328 eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
1330 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1331 eth_dev->data->port_id);
1332 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1336 /* Only support 1:1 queue/interrupt mapping so far.
1337 * TODO: support n:1 queue/interrupt mapping when there are limited number of
1338 * interrupt vectors (<N+1).
1341 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1344 struct virtio_hw *hw = dev->data->dev_private;
1346 PMD_INIT_LOG(INFO, "queue/interrupt binding");
1347 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1348 dev->intr_handle->intr_vec[i] = i + 1;
1349 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1350 VIRTIO_MSI_NO_VECTOR) {
1351 PMD_DRV_LOG(ERR, "failed to set queue vector");
1360 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1363 struct virtio_hw *hw = dev->data->dev_private;
1365 PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1366 for (i = 0; i < dev->data->nb_rx_queues; ++i)
1367 VTPCI_OPS(hw)->set_queue_irq(hw,
1368 hw->vqs[i * VTNET_CQ],
1369 VIRTIO_MSI_NO_VECTOR);
1373 virtio_configure_intr(struct rte_eth_dev *dev)
1375 struct virtio_hw *hw = dev->data->dev_private;
1377 if (!rte_intr_cap_multiple(dev->intr_handle)) {
1378 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1382 if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1383 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1387 if (!dev->intr_handle->intr_vec) {
1388 dev->intr_handle->intr_vec =
1389 rte_zmalloc("intr_vec",
1390 hw->max_queue_pairs * sizeof(int), 0);
1391 if (!dev->intr_handle->intr_vec) {
1392 PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1393 hw->max_queue_pairs);
1398 /* Re-register callback to update max_intr */
1399 rte_intr_callback_unregister(dev->intr_handle,
1400 virtio_interrupt_handler,
1402 rte_intr_callback_register(dev->intr_handle,
1403 virtio_interrupt_handler,
1406 /* DO NOT try to remove this! This function will enable msix, or QEMU
1407 * will encounter SIGSEGV when DRIVER_OK is sent.
1408 * And for legacy devices, this should be done before queue/vec binding
1409 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1410 * (22) will be ignored.
1412 if (virtio_intr_enable(dev) < 0) {
1413 PMD_DRV_LOG(ERR, "interrupt enable failed");
1417 if (virtio_queues_bind_intr(dev) < 0) {
1418 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1425 /* reset device and renegotiate features if needed */
1427 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1429 struct virtio_hw *hw = eth_dev->data->dev_private;
1430 struct virtio_net_config *config;
1431 struct virtio_net_config local_config;
1432 struct rte_pci_device *pci_dev = NULL;
1435 /* Reset the device although not necessary at startup */
1439 virtio_dev_free_mbufs(eth_dev);
1440 virtio_free_queues(hw);
1443 /* Tell the host we've noticed this device. */
1444 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1446 /* Tell the host we've known how to drive the device. */
1447 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1448 if (virtio_negotiate_features(hw, req_features) < 0)
1451 if (!hw->virtio_user_dev) {
1452 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1453 rte_eth_copy_pci_info(eth_dev, pci_dev);
1456 /* If host does not support both status and MSI-X then disable LSC */
1457 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1458 hw->use_msix != VIRTIO_MSIX_NONE)
1459 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1461 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1463 /* Setting up rx_header size for the device */
1464 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1465 vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1466 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1468 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1470 /* Copy the permanent MAC address to: virtio_hw */
1471 virtio_get_hwaddr(hw);
1472 ether_addr_copy((struct ether_addr *) hw->mac_addr,
1473 ð_dev->data->mac_addrs[0]);
1475 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1476 hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1477 hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1479 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1480 config = &local_config;
1482 vtpci_read_dev_config(hw,
1483 offsetof(struct virtio_net_config, mac),
1484 &config->mac, sizeof(config->mac));
1486 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1487 vtpci_read_dev_config(hw,
1488 offsetof(struct virtio_net_config, status),
1489 &config->status, sizeof(config->status));
1492 "VIRTIO_NET_F_STATUS is not supported");
1496 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1497 vtpci_read_dev_config(hw,
1498 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1499 &config->max_virtqueue_pairs,
1500 sizeof(config->max_virtqueue_pairs));
1503 "VIRTIO_NET_F_MQ is not supported");
1504 config->max_virtqueue_pairs = 1;
1507 hw->max_queue_pairs = config->max_virtqueue_pairs;
1509 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1510 vtpci_read_dev_config(hw,
1511 offsetof(struct virtio_net_config, mtu),
1513 sizeof(config->mtu));
1516 * MTU value has already been checked at negotiation
1517 * time, but check again in case it has changed since
1518 * then, which should not happen.
1520 if (config->mtu < ETHER_MIN_MTU) {
1521 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1526 hw->max_mtu = config->mtu;
1527 /* Set initial MTU to maximum one supported by vhost */
1528 eth_dev->data->mtu = config->mtu;
1531 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1532 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1535 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1536 config->max_virtqueue_pairs);
1537 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1539 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1540 config->mac[0], config->mac[1],
1541 config->mac[2], config->mac[3],
1542 config->mac[4], config->mac[5]);
1544 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1545 hw->max_queue_pairs = 1;
1546 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1547 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1550 ret = virtio_alloc_queues(eth_dev);
1554 if (eth_dev->data->dev_conf.intr_conf.rxq) {
1555 if (virtio_configure_intr(eth_dev) < 0) {
1556 PMD_INIT_LOG(ERR, "failed to configure interrupt");
1561 vtpci_reinit_complete(hw);
1564 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1565 eth_dev->data->port_id, pci_dev->id.vendor_id,
1566 pci_dev->id.device_id);
1572 * Remap the PCI device again (IO port map for legacy device and
1573 * memory map for modern device), so that the secondary process
1574 * could have the PCI initiated correctly.
1577 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1581 * We don't have to re-parse the PCI config space, since
1582 * rte_pci_map_device() makes sure the mapped address
1583 * in secondary process would equal to the one mapped in
1584 * the primary process: error will be returned if that
1585 * requirement is not met.
1587 * That said, we could simply reuse all cap pointers
1588 * (such as dev_cfg, common_cfg, etc.) parsed from the
1589 * primary process, which is stored in shared memory.
1591 if (rte_pci_map_device(pci_dev)) {
1592 PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1596 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1604 virtio_set_vtpci_ops(struct virtio_hw *hw)
1606 #ifdef RTE_VIRTIO_USER
1607 if (hw->virtio_user_dev)
1608 VTPCI_OPS(hw) = &virtio_user_ops;
1612 VTPCI_OPS(hw) = &modern_ops;
1614 VTPCI_OPS(hw) = &legacy_ops;
1618 * This function is based on probe() function in virtio_pci.c
1619 * It returns 0 on success.
1622 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1624 struct virtio_hw *hw = eth_dev->data->dev_private;
1627 RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1629 eth_dev->dev_ops = &virtio_eth_dev_ops;
1631 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1632 if (!hw->virtio_user_dev) {
1633 ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1638 virtio_set_vtpci_ops(hw);
1639 set_rxtx_funcs(eth_dev);
1644 /* Allocate memory for storing MAC addresses */
1645 eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1646 if (eth_dev->data->mac_addrs == NULL) {
1648 "Failed to allocate %d bytes needed to store MAC addresses",
1649 VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1653 hw->port_id = eth_dev->data->port_id;
1654 /* For virtio_user case the hw->virtio_user_dev is populated by
1655 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1657 if (!hw->virtio_user_dev) {
1658 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1663 /* reset device and negotiate default features */
1664 ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1668 /* Setup interrupt callback */
1669 if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1670 rte_intr_callback_register(eth_dev->intr_handle,
1671 virtio_interrupt_handler, eth_dev);
1676 rte_free(eth_dev->data->mac_addrs);
1681 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1683 PMD_INIT_FUNC_TRACE();
1685 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1688 virtio_dev_stop(eth_dev);
1689 virtio_dev_close(eth_dev);
1691 eth_dev->dev_ops = NULL;
1692 eth_dev->tx_pkt_burst = NULL;
1693 eth_dev->rx_pkt_burst = NULL;
1695 rte_free(eth_dev->data->mac_addrs);
1696 eth_dev->data->mac_addrs = NULL;
1698 /* reset interrupt callback */
1699 if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1700 rte_intr_callback_unregister(eth_dev->intr_handle,
1701 virtio_interrupt_handler,
1703 if (eth_dev->device)
1704 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1706 PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1711 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1712 struct rte_pci_device *pci_dev)
1714 return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1715 eth_virtio_dev_init);
1718 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1720 return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1723 static struct rte_pci_driver rte_virtio_pmd = {
1725 .name = "net_virtio",
1727 .id_table = pci_id_virtio_map,
1729 .probe = eth_virtio_pci_probe,
1730 .remove = eth_virtio_pci_remove,
1733 RTE_INIT(rte_virtio_pmd_init);
1735 rte_virtio_pmd_init(void)
1737 if (rte_eal_iopl_init() != 0) {
1738 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1742 rte_pci_register(&rte_virtio_pmd);
1746 * Configure virtio device
1747 * It returns 0 on success.
1750 virtio_dev_configure(struct rte_eth_dev *dev)
1752 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1753 struct virtio_hw *hw = dev->data->dev_private;
1754 uint64_t req_features;
1757 PMD_INIT_LOG(DEBUG, "configure");
1758 req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1760 if (dev->data->dev_conf.intr_conf.rxq) {
1761 ret = virtio_init_device(dev, hw->req_guest_features);
1766 /* The name hw_ip_checksum is a bit confusing since it can be
1767 * set by the application to request L3 and/or L4 checksums. In
1768 * case of virtio, only L4 checksum is supported.
1770 if (rxmode->hw_ip_checksum)
1771 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1773 if (rxmode->enable_lro)
1775 (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1776 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
1778 /* if request features changed, reinit the device */
1779 if (req_features != hw->req_guest_features) {
1780 ret = virtio_init_device(dev, req_features);
1785 if (rxmode->hw_ip_checksum &&
1786 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1788 "rx checksum not available on this host");
1792 if (rxmode->enable_lro &&
1793 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1794 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
1796 "Large Receive Offload not available on this host");
1800 /* start control queue */
1801 if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1802 virtio_dev_cq_start(dev);
1804 hw->vlan_strip = rxmode->hw_vlan_strip;
1806 if (rxmode->hw_vlan_filter
1807 && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1809 "vlan filtering not available on this host");
1813 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1814 /* Enable vector (0) for Link State Intrerrupt */
1815 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1816 VIRTIO_MSI_NO_VECTOR) {
1817 PMD_DRV_LOG(ERR, "failed to set config vector");
1821 rte_spinlock_init(&hw->state_lock);
1823 hw->use_simple_rx = 1;
1824 hw->use_simple_tx = 1;
1826 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
1827 if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
1828 hw->use_simple_rx = 0;
1829 hw->use_simple_tx = 0;
1832 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1833 hw->use_simple_rx = 0;
1834 hw->use_simple_tx = 0;
1837 if (rxmode->hw_ip_checksum)
1838 hw->use_simple_rx = 0;
1845 virtio_dev_start(struct rte_eth_dev *dev)
1847 uint16_t nb_queues, i;
1848 struct virtnet_rx *rxvq;
1849 struct virtnet_tx *txvq __rte_unused;
1850 struct virtio_hw *hw = dev->data->dev_private;
1853 /* Finish the initialization of the queues */
1854 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1855 ret = virtio_dev_rx_queue_setup_finish(dev, i);
1859 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1860 ret = virtio_dev_tx_queue_setup_finish(dev, i);
1865 /* check if lsc interrupt feature is enabled */
1866 if (dev->data->dev_conf.intr_conf.lsc) {
1867 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1868 PMD_DRV_LOG(ERR, "link status not supported by host");
1873 /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1874 * in device configure, but it could be unmapped when device is
1877 if (dev->data->dev_conf.intr_conf.lsc ||
1878 dev->data->dev_conf.intr_conf.rxq) {
1879 virtio_intr_disable(dev);
1881 if (virtio_intr_enable(dev) < 0) {
1882 PMD_DRV_LOG(ERR, "interrupt enable failed");
1887 /*Notify the backend
1888 *Otherwise the tap backend might already stop its queue due to fullness.
1889 *vhost backend will have no chance to be waked up
1891 nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1892 if (hw->max_queue_pairs > 1) {
1893 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1897 PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1899 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1900 rxvq = dev->data->rx_queues[i];
1901 /* Flush the old packets */
1902 virtqueue_rxvq_flush(rxvq->vq);
1903 virtqueue_notify(rxvq->vq);
1906 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1907 txvq = dev->data->tx_queues[i];
1908 virtqueue_notify(txvq->vq);
1911 PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1913 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1914 rxvq = dev->data->rx_queues[i];
1915 VIRTQUEUE_DUMP(rxvq->vq);
1918 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1919 txvq = dev->data->tx_queues[i];
1920 VIRTQUEUE_DUMP(txvq->vq);
1923 set_rxtx_funcs(dev);
1926 /* Initialize Link state */
1927 virtio_dev_link_update(dev, 0);
1932 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1934 struct virtio_hw *hw = dev->data->dev_private;
1935 uint16_t nr_vq = virtio_get_nr_vq(hw);
1936 const char *type __rte_unused;
1937 unsigned int i, mbuf_num = 0;
1938 struct virtqueue *vq;
1939 struct rte_mbuf *buf;
1942 if (hw->vqs == NULL)
1945 for (i = 0; i < nr_vq; i++) {
1950 queue_type = virtio_get_queue_type(hw, i);
1951 if (queue_type == VTNET_RQ)
1953 else if (queue_type == VTNET_TQ)
1959 "Before freeing %s[%d] used and unused buf",
1963 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
1964 rte_pktmbuf_free(buf);
1969 "After freeing %s[%d] used and unused buf",
1974 PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
1978 * Stop device: disable interrupt and mark link down
1981 virtio_dev_stop(struct rte_eth_dev *dev)
1983 struct virtio_hw *hw = dev->data->dev_private;
1984 struct rte_eth_link link;
1985 struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
1987 PMD_INIT_LOG(DEBUG, "stop");
1989 rte_spinlock_lock(&hw->state_lock);
1990 if (intr_conf->lsc || intr_conf->rxq)
1991 virtio_intr_disable(dev);
1994 memset(&link, 0, sizeof(link));
1995 rte_eth_linkstatus_set(dev, &link);
1996 rte_spinlock_unlock(&hw->state_lock);
2000 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2002 struct rte_eth_link link;
2004 struct virtio_hw *hw = dev->data->dev_private;
2006 memset(&link, 0, sizeof(link));
2007 link.link_duplex = ETH_LINK_FULL_DUPLEX;
2008 link.link_speed = ETH_SPEED_NUM_10G;
2009 link.link_autoneg = ETH_LINK_FIXED;
2011 if (hw->started == 0) {
2012 link.link_status = ETH_LINK_DOWN;
2013 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2014 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2015 vtpci_read_dev_config(hw,
2016 offsetof(struct virtio_net_config, status),
2017 &status, sizeof(status));
2018 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2019 link.link_status = ETH_LINK_DOWN;
2020 PMD_INIT_LOG(DEBUG, "Port %d is down",
2021 dev->data->port_id);
2023 link.link_status = ETH_LINK_UP;
2024 PMD_INIT_LOG(DEBUG, "Port %d is up",
2025 dev->data->port_id);
2028 link.link_status = ETH_LINK_UP;
2031 return rte_eth_linkstatus_set(dev, &link);
2035 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2037 const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2038 struct virtio_hw *hw = dev->data->dev_private;
2040 if (mask & ETH_VLAN_FILTER_MASK) {
2041 if (rxmode->hw_vlan_filter &&
2042 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2045 "vlan filtering not available on this host");
2051 if (mask & ETH_VLAN_STRIP_MASK)
2052 hw->vlan_strip = rxmode->hw_vlan_strip;
2058 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2060 uint64_t tso_mask, host_features;
2061 struct virtio_hw *hw = dev->data->dev_private;
2063 dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2065 dev_info->pci_dev = dev->device ? RTE_ETH_DEV_TO_PCI(dev) : NULL;
2066 dev_info->max_rx_queues =
2067 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2068 dev_info->max_tx_queues =
2069 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2070 dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2071 dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2072 dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2073 dev_info->default_txconf = (struct rte_eth_txconf) {
2074 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
2077 host_features = VTPCI_OPS(hw)->get_features(hw);
2078 dev_info->rx_offload_capa = 0;
2079 if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2080 dev_info->rx_offload_capa |=
2081 DEV_RX_OFFLOAD_TCP_CKSUM |
2082 DEV_RX_OFFLOAD_UDP_CKSUM;
2084 tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2085 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2086 if ((host_features & tso_mask) == tso_mask)
2087 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2089 dev_info->tx_offload_capa = 0;
2090 if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2091 dev_info->tx_offload_capa |=
2092 DEV_TX_OFFLOAD_UDP_CKSUM |
2093 DEV_TX_OFFLOAD_TCP_CKSUM;
2095 tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2096 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2097 if ((hw->guest_features & tso_mask) == tso_mask)
2098 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2102 * It enables testpmd to collect per queue stats.
2105 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2106 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2107 __rte_unused uint8_t is_rx)
2112 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2113 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2114 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2116 RTE_INIT(virtio_init_log);
2118 virtio_init_log(void)
2120 virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2121 if (virtio_logtype_init >= 0)
2122 rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2123 virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2124 if (virtio_logtype_driver >= 0)
2125 rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);